Hi, im performing analysis using Logistic regression and GEE, where all my variables are categorical, i need to test for multicollinearity among the variables. What test to do and how to perform it in either SPSS or Stata?
if I'm not mistaken Chapter 7 , 10 and 11 of Applied Linear Statistical Models book written by John Neter (Fifth Edition) addresses multicollinearity. You can study more about that.
but in the SPSS as i know do the steps :
Analyse > Regression > Linear and then add your dependent and independent variables in the specified box and then under the tab Statistics check Collinearity Diagnostics
in the output you will have values for checking the multicollinearity.
Thank you David and Samira, i have seen some people post that VIF cant be used for multicollinearity assessment for categorical variables, any comments on that? As i cant decide in which test to use
In linear regression, each category of a categorical conceptual variable becomes a separate predictor variable (usually, a 1-0 "dummy" variable), with one category omitted from the model. You CAN ignore the categorical nature of the conceptual variable and obtain a VIF value for each separate predictor (dummy) variable. Those VIF values are legitimate and interpretable in the usual way. However, the VIF values you obtain will depend on which category you arbitrarily choose to omit from the model. Omit a different category, and the VIF values for the other categories will change. That makes for ambiguity in interpretation, since when the values differ the substantive interpretations can also differ depending on which category is omitted.
The Generalized Variance Inflation Factor (GVIF) tries to solve that problem, by calculating a VIF-style statistic for each SUBSET of predictors, where a subset includes the categories comprising the conceptual variable. You can search online for sources about the GVIF, some of which are mentioned in a related
thread on ResearchGate: https://www.researchgate.net/post/Anyone_familiar_with_VIF_Variance_Inflation_Factor_and_categorical_variables